codetether_agent/session/tasks/event/
goal_source.rs

1//! Goal provenance categories.
2//!
3//! This module defines the trust category attached to a session goal when it is
4//! written to the append-only task log. The category records where the goal came
5//! from, allowing recovery, governance, and debugging code to distinguish goals
6//! grounded in raw transcript evidence from goals inferred through weaker or
7//! lossy sources.
8//!
9//! The enum is serialized as `snake_case` so goal provenance remains stable and
10//! readable in persisted JSONL task logs.
11
12use serde::{Deserialize, Serialize};
13
14/// Provenance category for a declared session goal.
15///
16/// `GoalSourceKind` describes the evidence source used to create a session
17/// objective. Consumers should treat variants as a rough trust hierarchy:
18/// direct raw-turn or user-provided goals are stronger than summarized recall,
19/// memory-derived, or inferred goals.
20///
21/// The default is [`GoalSourceKind::Inferred`], which is intentionally the
22/// weakest category. This keeps older log entries and omitted provenance fields
23/// safe by marking them as unverified rather than pretending they came from a
24/// canonical user turn.
25#[derive(Clone, Debug, Default, PartialEq, Eq, Serialize, Deserialize)]
26#[serde(rename_all = "snake_case")]
27pub enum GoalSourceKind {
28    /// The goal was anchored to a raw transcript turn.
29    ///
30    /// Use this when the objective is derived directly from canonical session
31    /// history, such as a specific user message returned by `context_browse`.
32    RawTurn,
33
34    /// The goal was explicitly provided by the user.
35    ///
36    /// This is appropriate for direct goal-setting interfaces such as a TUI
37    /// `/goal set ...` command or a user request that intentionally declares the
38    /// current objective.
39    UserProvided,
40
41    /// The goal was derived from summarized session recall.
42    ///
43    /// Recall summaries are useful for recovery but may omit, blend, or
44    /// mis-rank details from raw history. Consumers should verify this kind
45    /// against transcript turns before treating it as authoritative.
46    RecallSummary,
47
48    /// The goal was derived from curated durable memory.
49    ///
50    /// Memory can preserve important decisions across sessions, but it may be
51    /// stale or broader than the current conversation. Consumers should consider
52    /// memory scope and timestamp before using it as a hard constraint.
53    Memory,
54
55    /// The goal was inferred without stronger provenance.
56    ///
57    /// This is the fallback for missing provenance and should be treated as
58    /// low-confidence until confirmed by the user or anchored to raw transcript
59    /// evidence.
60    #[default]
61    Inferred,
62}
codetether_agent/session/tasks/event/goal_source.rs

codetether_agent/session/tasks/event/
goal_source.rs
